fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-08 11:18:08 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

i965/fs: Lower 32-wide scratch writes in the generator.

Browse source 
The hardware has messages that can write 32 32bit components at once
but the channel enable mask gets messed up.  We need to split them
into several 16-wide scratch writes for the channel enables to be
applied correctly.  The SIMD lowering pass cannot be used for this
because scratch writes are emitted rather late during register
allocation long after SIMD lowering has been done.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This commit is contained in:
Francisco Jerez 2016-05-27 23:29:02 -07:00
parent a7d319c00b
commit 1e3c58ffaf
1 changed files with 24 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
src/mesa/drivers/dri/i965/brw_fs_generator.cpp
View file

@ -1061,14 +1061,32 @@ fs_generator::generate_discard_jump(fs_inst *inst)
void
fs_generator::generate_scratch_write(fs_inst *inst, struct brw_reg src)
{
assert(inst->exec_size <= 16 || inst->force_writemask_all);
/* The 32-wide messages only respect the first 16-wide half of the channel
* enable signals which are replicated identically for the second group of
* 16 channels, so we cannot use them unless the write is marked
* force_writemask_all.
*/
const unsigned lower_size = inst->force_writemask_all ? inst->exec_size :
MIN2(16, inst->exec_size);
const unsigned block_size = 4 * lower_size / REG_SIZE;
assert(inst->mlen != 0);
brw_MOV(p,
brw_uvec_mrf(inst->exec_size, (inst->base_mrf + 1), 0),
retype(src, BRW_REGISTER_TYPE_UD));
brw_oword_block_write_scratch(p, brw_message_reg(inst->base_mrf),
inst->exec_size / 8, inst->offset);
brw_push_insn_state(p);
brw_set_default_exec_size(p, cvt(lower_size) - 1);
brw_set_default_compression(p, lower_size > 8);
for (unsigned i = 0; i < inst->exec_size / lower_size; i++) {
brw_set_default_group(p, (inst->force_sechalf ? 8 : 0) + lower_size * i);
brw_MOV(p, brw_uvec_mrf(lower_size, inst->base_mrf + 1, 0),
retype(offset(src, block_size * i), BRW_REGISTER_TYPE_UD));
brw_oword_block_write_scratch(p, brw_message_reg(inst->base_mrf),
block_size,
inst->offset + block_size * REG_SIZE * i);
}
brw_pop_insn_state(p);
}
void